Nvidia Shows and Demos Testla Volta V100 has 5120 Shader processors

People are confusing Tesla which is more of a data server thing with nvidia's usual GTX graphic card lineup. Also the Tesla P-100 (pascal based) already has HBM2 and already available for about 5 grand so them using HBM2 is not a new thing and doesn't mean Volta based GTX will use it.

 

Wait what? 1080 was GP104.

 

How can you say that? You forgot how much nvidia asked for Pascal FE cards when it came out? You will be paying 500$ for the 2070.

 

He's saying that if you take the size cut from GP100 to GP104 and applied it to GV100, it would still be bigger than GP102. The problem is GV100 is a completely new architecture and doing that comparison is not going to be accurate at all.

 

That's why Polaris is great at just 232mm^2.
The smaller the chip the more they can fit in a single wafer and make it more affordable.
Having high performance is great but it would be awesome if everyone could afford it!

They are already expensive af.

Even if they can use 90% of the dies, it will have less dies per wafer. So each die will be more expensive. 800mm^2 is completely nuts.

I'm sure the full Volta GPU will cost about $1499.
2080 will cost about $999.
2070 $500? nah...

That die size could be bad, if they had any competition. AMD will take 1 or 2 years to make something after Vega.

 

Thanks Hilbert, I missed the fact that the 15 TFLOPS are included/written in the slide!

It's a bit disappointing that this massive 800mm squared die is 'only' has about 50% more processing power than the GTX 1080ti. Other people have mentioned how expensive a gaming version of this product would be due to the large die size. Now 50% increase in performance wouldn't be bad, but it might be unlikely we'll get such a large chip given the large price that would go with it, so maybe Volta won't be that much of an improvement over Pascal. Although, I'm sure I saw NVidia slides showing twice the performance of Volta over Pascal on their roadmap, but this large Testla chip is only 50% more processing performance. Hmmm....

Maybe, but as some people have been saying that 1180ti is gonna be very very expensive so might not happen this way!

 

The die size is for a datacenter oriented product. GP100 is a completely different design than GP102/104. Volta is clearly taking this to a new level. Cut the FP64, the new Tensor cores, etc and the chip is going to be like a 1/3rd the size.

Trying to extrapolate price from a die size comparison of a product that's completely separated from the gaming market is like really stretching it.

 

Question. Does the 800mm^2 include HBM?

As Denial says GTX die will be much smaller. GP100 is 610mm^2 and GP102 is 471mm^2. Full size GTX Volta may be more around 560mm^2. Which is not a huge jump from GP102. Vega is rumored around 520mm^2 for reference and the users in here talking about how expensive the Volta cards will be based on die size think Vega will be retailing for $300.

 

show me the next xx60 card!

 

According to PC Perspective the max interposer size is 830mm2.

https://www.pcper.com/reviews/Graph...ls-Tesla-P100-GP100-GPU/Joshs-Thoughts-P100-1

But the article is worded like GP100's 610mm2 is just the die only. So I'm not sure.

 

https://devblogs.nvidia.com/parallelforall/wp-content/uploads/2017/05/image3.png

Apparently the die mm2 includes the HBM2/Interposer. So the actual die is smaller. Also the tensor units are why FP32 is low given the die size itself. The tensor units most likely won't be on the gaming cards - or will be severely cut down. Along with the FP64.

Edit: I guess I could have appended this to my previous post. Sorry.

 

Nerdgasmic. I want one with hacked drivers :nerd:

 

So ~15Tflop for GV100, midget high-end GV104 ~ 12Tflop?

Yeah, I was about to say its probably with HBM interposer,..


Im sure normal GTX variant will remain in the same price range 699$, 1080Ti settled to ~700- 750€, just like 980Ti after 1-2months.
https://geizhals.eu/?cat=gra16_512&xf=9810_7+10609+-+GTX+1080+Ti&sort=p

 

The big news I'm getting out of this is that Volta has had a third process switch and it's now again at 16/12nm. Which means it will most likely come for consumer products earlier than the end of Q1 2018, as initially predicted.

The gaming teraflop numbers aren't exactly crazy though, an overclocked 1080Ti will get close to these. That's pretty much a 2GHz 1080Ti. It's not as huge a leap as I imagined it to be, so I'm curious what different things the actual architecture holds. They seem to have gone very wide this time, I wonder if other architectural details are different or this is "just" a wider Pascal.

Interesting times ahead. And yes, that chip is massive.

About the die size, isn't Vega 530-560mm2 and 12.5Tflop? What I'm reading wrong here? Are the tensor parts that much of the die size? It seems like they aren't, considering that the increase of SM from 56-80 seems to be almost proportional to the increase in die size.

 

If these clock similar to Pascal however we may see 18.5-19 Tflop performance with this chip. Also as previously established the die size discussed includes the HBM imposters.

To your edit. The doubled the tensor core count from GP100. As fare as I know they were not a part of GP102.

Does AMD include HBM impossers in their die measurement as I would estimate GV100 at 580~610mm^2 without HBM.

 

https://devblogs.nvidia.com/parallelforall/inside-volta/?ncid=so-twi-vt-13918

Has some more information about architectural changes.

Seems like PRMinister was right - they kind of copied AMD's implementation of thread scheduling although it isn't clear how similar it is to what AMD does. Probably won't find out till closer to launch.

 

Its hard to conclude anything about GTX line from this GV100 data. Two being completely bifurcated and GV100 carrying significant amount of extra weight. One of few things they do share is:

Volta SM (Streaming Multiprocessor)

• New mixed-precision FP16/FP32 Tensor Cores purpose-built for deep learning matrix arithmetic;
• Enhanced L1 data cache for higher performance and lower latency;
• Streamlined instruction set for simpler decoding and reduced instruction latencies
• higher clocks and higher power efficiency.

https://devblogs.nvidia.com/parallelforall/inside-volta/?ncid=so-twi-vt-13918

 

These are apparently the key new features of GV100:

• New Streaming Multiprocessor (SM) Architecture Optimized for Deep Learning
  • Second-Generation NVLink™
  • HBM2 Memory: Faster, Higher Efficiency
  • Volta Multi-Process Service
  • Enhanced Unified Memory and Address Translation Services
  • Cooperative Groups and New Cooperative Launch APIs
  • Maximum Performance and Maximum Efficiency Modes
  • Volta Optimized Software
  
  Out of all of that, almost zero seems to deal with new deeper architectural stuff. They pretty much seem to have a more optimized/refined memory controller and that's it. The Maximum Performance and Maximum Efficiency modes remind me of similar wording from AMD about Hawaii and Fiji. I don't believe that this one will clock that high this time. We'll see.

 

• 
  
  
  Something in the architecture changed to enable a 50% increase in efficiency.

 

I'll get excited when the software catches up to the hardware. I've fallen for the hype of hardware too many times, to just sit on it playing console ports. Sounds impressive, but until I see a game that NEEDS it, I'm not impressed. Why buy a V8 if you're only driving to work?